vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

style : minor

Browse source
This commit is contained in:
Georgi Gerganov 2024-06-06 10:21:35 +03:00
parent 7ab6023bc3
commit 4c4d877d23
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
3 changed files with 4 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

2
gguf-py/gguf/constants.py
View file

@ -415,6 +415,7 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.TOKEN_EMBD, MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.TOKEN_EMBD_NORM, MODEL_TENSOR.TOKEN_EMBD_NORM,
MODEL_TENSOR.TOKEN_TYPES, MODEL_TENSOR.TOKEN_TYPES,
MODEL_TENSOR.ATTN_NORM_2,
MODEL_TENSOR.ATTN_OUT_NORM, MODEL_TENSOR.ATTN_OUT_NORM,
MODEL_TENSOR.ATTN_Q, MODEL_TENSOR.ATTN_Q,
MODEL_TENSOR.ATTN_Q_NORM, MODEL_TENSOR.ATTN_Q_NORM,
@ -426,7 +427,6 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_GATE, MODEL_TENSOR.FFN_GATE,
MODEL_TENSOR.FFN_DOWN, MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.LAYER_OUT_NORM, MODEL_TENSOR.LAYER_OUT_NORM,
MODEL_TENSOR.ATTN_NORM_2,
], ],
MODEL_ARCH.MPT: [ MODEL_ARCH.MPT: [
MODEL_TENSOR.TOKEN_EMBD, MODEL_TENSOR.TOKEN_EMBD,

5
llama.cpp
View file

@ -704,6 +704,7 @@ static const std::map<llm_arch, std::map<llm_tensor, std::string>> LLM_TENSOR_NA
{ LLM_TENSOR_TOKEN_EMBD, "token_embd" }, { LLM_TENSOR_TOKEN_EMBD, "token_embd" },
{ LLM_TENSOR_TOKEN_EMBD_NORM, "token_embd_norm" }, { LLM_TENSOR_TOKEN_EMBD_NORM, "token_embd_norm" },
{ LLM_TENSOR_TOKEN_TYPES, "token_types" }, { LLM_TENSOR_TOKEN_TYPES, "token_types" },
{ LLM_TENSOR_ATTN_NORM_2, "blk.%d.attn_norm_2" },
{ LLM_TENSOR_ATTN_OUT_NORM, "blk.%d.attn_output_norm" }, { LLM_TENSOR_ATTN_OUT_NORM, "blk.%d.attn_output_norm" },
{ LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" },
{ LLM_TENSOR_ATTN_Q_NORM, "blk.%d.attn_q_norm" }, { LLM_TENSOR_ATTN_Q_NORM, "blk.%d.attn_q_norm" },
@ -715,7 +716,6 @@ static const std::map<llm_arch, std::map<llm_tensor, std::string>> LLM_TENSOR_NA
{ LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" },
{ LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" },
{ LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" },
{ LLM_TENSOR_ATTN_NORM_2, "blk.%d.attn_norm_2" },
}, },
}, },
{ {
@ -8524,8 +8524,7 @@ struct llm_build_context {
cur = llm_build_norm(ctx0, cur, hparams, model.layers[il].attn_out_norm, model.layers[il].attn_out_norm_b, LLM_NORM, cb, il); cur = llm_build_norm(ctx0, cur, hparams, model.layers[il].attn_out_norm, model.layers[il].attn_out_norm_b, LLM_NORM, cb, il);
if (model.layers[il].attn_norm_2 != nullptr) { if (model.layers[il].attn_norm_2 != nullptr) {
// re-add the layer input cur = ggml_add(ctx0, cur, inpL); // re-add the layer input
cur = ggml_add(ctx0, cur, inpL);
cur = llm_build_norm(ctx0, cur, hparams, model.layers[il].attn_norm_2, model.layers[il].attn_norm_2_b, LLM_NORM, cb, il); cur = llm_build_norm(ctx0, cur, hparams, model.layers[il].attn_norm_2, model.layers[il].attn_norm_2_b, LLM_NORM, cb, il);
} }